Training β-VAE by Aggregating a Learned Gaussian Posterior with a Decoupled Decoder

The reconstruction loss and the Kullback-Leibler divergence (KLD) loss in a variational autoencoder (VAE) often play antagonistic roles, and tuning the weight of the KLD loss in beta-VAE to achieve a balance between the two losses is a tricky and dataset-specific task. As a result, current practices in VAE training often result in a trade-off between the reconstruction fidelity and the continuity/disentanglement of the latent space, if the weight beta is not carefully tuned. In this paper, we present intuitions and a careful analysis of the antagonistic mechanism of the two losses, and propose, based on the insights, a simple yet effective two-stage method for training a VAE. Specifically, the method aggregates a learned Gaussian posterior z similar to q.(z| x) with a decoder decoupled from the KLD loss, which is trained to learn a new conditional distribution pf(x|z) of the input data x. Experimentally, we show that the aggregated VAE maximally satisfies the Gaussian assumption about the latent space, while still achieves a reconstruction error comparable to when the latent space is only loosely regularized by N(0, I). The proposed approach does not require hyperparameter (i.e., the KLD weight beta) tuning given a specific dataset as required in common VAE training practices. We evaluate the method using a medical dataset intended for 3D skull reconstruction and shape completion, and the results indicate promising generative capabilities of the VAE trained using the proposed method. Besides, through guided manipulation of the latent variables, we establish a connection between existing autoencoder (AE)-based approaches and generative approaches, such as VAE, for the shape completion problem. Codes and pre-trained weights are available at https://github.com/ Jianningli/skullVAE.

Automated summary

The reconstruction loss and the Kullback-Leibler divergence (KLD) loss in a variational autoencoder (VAE) often have opposing effects, making it difficult to achieve a balance between the two. This paper proposes a two-stage method for training a VAE that solves this problem by decoupling the KLD loss from the decoder. Experimental results show that the proposed method achieves a good balance between the Gaussian assumption of the latent space and reconstruction error, without requiring specific tuning of hyperparameters. The method is evaluated using a medical dataset for skull reconstruction and shape completion, demonstrating promising generative capabilities.